Perturbation Methods II: General Case

(Lectures on Solution Methods for Economists VI)

Jesús Fernández-Villaverde,1 Pablo Guerrón,2 and David Zarruk Valencia3

November 18, 2019
1
University of Pennsylvania
2
Boston College
3
ITAM
The General case

• Most of arguments in the previous set of lecture notes are easy to generalize.

• The set of equilibrium conditions of many DSGE models can be written using recursive notation as:

Et H(y , y 0 , x, x 0 ) = 0,

where yt is a ny × 1 vector of controls and xt is a nx × 1 vector of states.

• n = nx + ny .

• H maps R ny × R ny × R nx × R nx into R n .

1
Partitioning the state vector

• The state vector xt can be partitioned as x = [x1 ; x2 ]t .

• x1 is a (nx − n ) × 1 vector of endogenous state variables.

• x2 is a n × 1 vector of exogenous state variables.

• Why do we want to partition the state vector?

2
Exogenous stochastic process I

x20 = Ax2 + λη 0

• Process with 3 parts:

1. The deterministic component Ax2 , where A is a n × n matrix, with all eigenvalues with modulus less
than one.

2. The scaled innovation η 0 , where:

2.1 η is a known n × n matrix.

2.2  is a n × 1 i.i.d. innovation with bounded support, zero mean, and variance/covariance matrix I .

3. The perturbation parameter λ.

3
Exogenous stochastic process II

• We can accommodate very general structures of x2 through changes in the definition of the state
space: i.e. stochastic volatility.

• More general structure:

x20 = Γ(x2 ) + λη 0
where Γ is a non-linear function satisfying that all eigenvalues of its first derivative evaluated at the
non-stochastic steady state lie within the unit circle.

• Note we do not impose Gaussanity.

4
The perturbation parameter

• The scalar λ ≥ 0 is the perturbation parameter.

• If we set λ = 0, we have a deterministic model.

• Important: there is only ONE perturbation parameter. The matrix η takes account of relative sizes
of different shocks.

• Why bounded support? Samuelson (1970), Jin and Judd (2002).

5
Solution of the model

• The solution to the model is of the form:

y = g (x; λ)
x = h(x; λ) + λη0
0

where g maps R nx × R + into R ny and h maps R nx × R + into R nx .

• The matrix η is of order nx × n and is given by:

" #
∅
η=
η

6
Perturbation

• We wish to find a perturbation approximation of the functions g and h around the non-stochastic
steady state, xt = x̄ and λ = 0.

• We define the non-stochastic steady state as vectors (x̄, ȳ ) such that:

H(ȳ , ȳ , x̄, x̄) = 0.

• Note that ȳ = g (x̄; 0) and x̄ = h(x̄; 0).

• This is because, if λ = 0, Et H = H.

7
Plugging-in the proposed solution

• Substituting the proposed solution, we define:

F (x; λ) ≡ Et H(g (x; λ), g (h(x; λ) + ηλ0 , λ), x, h(x; λ) + ηλ0 ) = 0

• Since F (x; λ) = 0 for any values of x and λ, the derivatives of any order of F must also be equal to
zero.

• Formally:

Fx k λj (x; λ) = 0 ∀x, λ, j, k,

where Fx k λj (x, λ) denotes the derivative of F with respect to x taken k times and with respect to λ
taken j times. 8
First-order approximation

• We are looking for approximations to g and h around (x, λ) = (x̄, 0) of the form:
g (x; λ) = g (x̄; 0) + gx (x̄; 0)(x − x̄) + gλ (x̄; 0)λ
h(x; λ) = h(x̄; 0) + hx (x̄; 0)(x − x̄) + hλ (x̄; 0)λ

• As explained earlier, g (x̄; 0) = ȳ and h(x̄; 0) = x̄.

• The remaining four unknown coefficients of the first-order approximation to g and h are found by
using the fact that:
Fx (x̄; 0) = 0
and
Fλ (x̄; 0) = 0

• Before doing so, we need to introduce the tensor notation.

9
Tensors

• General trick from physics.

• An nth -rank tensor in a m-dimensional space is an operator that has n indices and mn components
and obeys certain transformation rules.

• [Hy ]iα is the (i, α) element of the derivative of H with respect to y :

1. The derivative of H with respect to y is an n × ny matrix.
2. Thus, [Hy ]iα is the element of this matrix located at the intersection of the i-th row and α-th column.
β Pny Pnx ∂Hi ∂g α ∂hβ
3. Thus, [Hy ]iα [gx ]α
β [hx ]j = α=1 β=1 ∂y α ∂x β ∂x j .

• [Hy 0 y 0 ]iαγ :
1. Hy 0 y 0 is a three dimensional array with n rows, ny columns, and ny pages.
2. Then [Hy 0 y 0 ]iαγ denotes the element of Hy 0 y 0 located at the intersection of row i, column α and page γ.

10
Solving the system I

• gx and hx can be found as the solution to the system:

β β
[Fx (x̄; 0)]ij = [Hy 0 ]iα [gx ]α i α i i
β [hx ]j + [Hy ]α [gx ]j + [Hx 0 ]β [hx ]j + [Hx ]j = 0;

i = 1, . . . , n; j, β = 1, . . . , nx ; α = 1, . . . , ny

• Note that the derivatives of H evaluated at (y , y 0 , x, x 0 ) = (ȳ , ȳ , x̄, x̄) are known.

• Then, we have a system of n × nx quadratic equations in the n × nx unknowns given by the elements
of gx and hx .

• We can solve with a standard quadratic matrix equation solver.

11
Solving the system II

• gλ and hλ are the solution to the n equations:

[Fλ (x̄; 0)]i =
β 0 φ
Et {[Hy 0 ]iα [gx ]α β i α i
β [hλ ] + [Hy 0 ]α [gx ]β [η]φ [ ] + [Hy 0 ]α [gλ ]
α

+[Hy ]iα [gλ ]α + [Hx 0 ]iβ [hλ ]β + [Hx 0 ]iβ [η]βφ [0 ]φ }
i = 1, . . . , n; α = 1, . . . , ny ; β = 1, . . . , nx ; φ = 1, . . . , n .
• Then: [Fλ (x̄; 0)]i
= [Hy 0 ]iα [gx ]α β i α i α i β
β [hλ ] + [Hy 0 ]α [gλ ] + [Hy ]α [gλ ] + [fx 0 ]β [hλ ] = 0;

i = 1, . . . , n; α = 1, . . . , ny ; β = 1, . . . , nx ; φ = 1, . . . , n .
• Certainty equivalence: linear and homogeneous equation in gλ and hλ . Thus, if a unique solution
exists, it satisfies:
hλ = 0
gλ = 0
12
 Second-order approximation I

The second-order approximations to g around (x; λ) = (x̄; 0) is

[g (x; λ)]i = [g (x̄; 0)]i + [gx (x̄; 0)]ia [(x − x̄)]a + [gλ (x̄; 0)]i [λ]
1
+ [gxx (x̄; 0)]iab [(x − x̄)]a [(x − x̄)]b
2
1
+ [gxλ (x̄; 0)]ia [(x − x̄)]a [λ]
2
1
+ [gλx (x̄; 0)]ia [(x − x̄)]a [λ]
2
1
+ [gλλ (x̄; 0)]i [λ][λ]
2
where i = 1, . . . , ny , a, b = 1, . . . , nx , and j = 1, . . . , nx .

13
 Second-order approximation II

The second-order approximations to h around (x; λ) = (x̄; 0) is

[h(x; λ)]j = [h(x̄; 0)]j + [hx (x̄; 0)]ja [(x − x̄)]a + [hλ (x̄; 0)]j [λ]
1
+ [hxx (x̄; 0)]jab [(x − x̄)]a [(x − x̄)]b
2
1
+ [hxλ (x̄; 0)]ja [(x − x̄)]a [λ]
2
1
+ [hλx (x̄; 0)]ja [(x − x̄)]a [λ]
2
1
+ [hλλ (x̄; 0)]j [λ][λ],
2
where i = 1, . . . , ny , a, b = 1, . . . , nx , and j = 1, . . . , nx .

14
Second-order approximation III

• The unknowns of these expansions are [gxx ]iab , [gxλ ]ia , [gλx ]ia , [gλλ ]i , [hxx ]jab , [hxλ ]ja , [hλx ]ja , [hλλ ]j .

• These coefficients can be identified by taking the derivative of F (x; λ) with respect to x and λ twice
and evaluating them at (x; λ) = (x̄; 0).

• By the arguments provided earlier, these derivatives must be zero.

15
 Solving the system I

We use Fxx (x̄; 0) to identify gxx (x̄; 0) and hxx (x̄; 0):

[Fxx (x̄; 0)]ijk =

[Hy 0 y 0 ]iαγ [gx ]γδ [hx ]δk + [Hy 0 y ]iαγ [gx ]γk + [Hy 0 x 0 ]iαδ [hx ]δk + [Hy 0 x ]iαk [gx ]α β


β [hx ]j
β β
+[Hy 0 ]iα [gxx ]α δ i α
βδ [hx ]k [hx ]j + [Hy 0 ]α [gx ]β [hxx ]jk

+ [Hyy 0 ]iαγ [gx ]γδ [hx ]δk + [Hyy ]iαγ [gx ]γk + [Hyx 0 ]iαδ [hx ]δk + [Hyx ]iαk [gx ]α


j

+[Hy ]iα [gxx ]α

jk

+ [Hx 0 y 0 ]iβγ [gx ]γδ [hx ]δk + [Hx 0 y ]iβγ [gx ]γk + [Hx 0 x 0 ]iβδ [hx ]δk + [Hx 0 x ]iβk [hx ]βj

+[Hx 0 ]iβ [hxx ]βjk

+[Hxy 0 ]ijγ [gx ]γδ [hx ]δk + [Hxy ]ijγ [gx ]γk + [Hxx 0 ]ijδ [hx ]δk + [Hxx ]ijk = 0;
i = 1, . . . n, j, k, β, δ = 1, . . . nx ; α, γ = 1, . . . ny .

16
Solving the system II

• We know the derivatives of H.

• We also know the first derivatives of g and h evaluated at (y , y 0 , x, x 0 ) = (ȳ , ȳ , x̄, x̄).

• Hence, the above expression represents a system of n × nx × nx linear equations in then n × nx × nx

unknowns elements of gxx and hxx .

17
 Solving the system III

Similarly, gλλ and hλλ can be obtained by solving:

[Fλλ (x̄; 0)]i = [Hy 0 ]iα [gx ]α

β [hλλ ]
β

+[Hy 0 y 0 ]iαγ [gx ]γδ [η]δξ [gx ]α β φ

β [η]φ [I ]ξ
β φ
+[Hy 0 x 0 ]iαδ [η]δξ [gx ]α
β [η]φ [I ]ξ
β φ
+[Hy 0 ]iα [gxx ]α δ i
βδ [η]ξ [η]φ [I ]ξ + [Hy 0 ]α [gλλ ]
α

+[Hy ]iα [gλλ ]α + [Hx 0 ]iβ [hλλ ]β

+[Hx 0 y 0 ]iβγ [gx ]γδ [η]δξ [η]βφ [I ]φξ
+[Hx 0 x 0 ]iβδ [η]δξ [η]βφ [I ]φξ = 0;
i = 1, . . . , n; α, γ = 1, . . . , ny ; β, δ = 1, . . . , nx ; φ, ξ = 1, . . . , n

a system of n linear equations in the n unknowns given by the elements of gλλ and hλλ .

18
Cross-derivatives

• The cross derivatives gxλ and hxλ are zero when evaluated at (x̄, 0).
• Why? Write the system Fλx (x̄; 0) = 0 taking into account that all terms containing either gλ or hλ
are zero at (x̄, 0).
• Then:
β γ
[Fλx (x̄; 0)]ij = [Hy 0 ]iα [gx ]α i α
β [hλx ]j + [Hy 0 ]α [gλx ]γ [hx ]j +
β
[Hy ]iα [gλx ]α i
j + [Hx 0 ]β [hλx ]j = 0;

i = 1, . . . n; α = 1, . . . , ny ; β, γ, j = 1, . . . , nx .
• This is a system of n × nx equations in the n × nx unknowns given by the elements of gλx and hλx .
• The system is homogeneous in the unknowns.
• Thus, if a unique solution exists, it is given by:
gλx = 0
hλx = 0
19
Structure of the solution

• The perturbation solution of the model satisfies:

gλ (x̄; 0) = 0
hλ (x̄; 0) = 0
gxλ (x̄; 0) = 0
hxλ (x̄; 0) = 0

• Standard deviation only appears in:

1. A constant term given by 12 gλλ λ2 for the control vector yt .
2. The first nx − n elements of 21 hλλ λ2 .
• Correction for risk.
• Quadratic terms in endogenous state vector x1 .
• Those terms capture non-linear behavior.
20
Higher-order approximations

• We can iterate this procedure as many times as we want.

• We can obtain n-th order approximations.

• Problems:

1. Existence of higher order derivatives (Santos, 1992).

2. Numerical instabilities.

3. Computational costs.

21